Re: [PATCH v8 03/12] virtio-sound: handle control messages and streams

[Philippe Mathieu-Daudé]

On 4/9/23 13:46, Manos Pitsidianakis wrote:
> On Mon, 04 Sep 2023 14:30, Philippe Mathieu-Daudé <philmd@linaro.org> 
> wrote:
> > On 4/9/23 13:00, Manos Pitsidianakis wrote:
> > > On Mon, 04 Sep 2023 13:46, Philippe Mathieu-Daudé <philmd@linaro.org> 
> > > wrote:
> > > > > +    size_t sz = iov_to_buf(cmd->elem->out_sg,
> > > > > +                           cmd->elem->out_num,
> > > > > +                           0,
> > > > > +                           &cmd->ctrl,
> > > > > +                           sizeof(cmd->ctrl));
> > > > > +    if (sz != sizeof(cmd->ctrl)) {
> > > > > +        qemu_log_mask(LOG_GUEST_ERROR,
> > > > > +                "%s: virtio-snd command size incorrect %zu vs \
> > > > > +                %zu\n", __func__, sz, sizeof(cmd->ctrl));
> > > > > +        return;
> > > > > +    }
> > > > > +
> > > > > +    trace_virtio_snd_handle_code(cmd->ctrl.code,
> > > >
> > > > IIUC the spec, this structure is in little endian, is that right?
> > > > So shouldn't swap various fields in this series?
> > >
> > > Not sure about the answer to this. Need input from someone more 
> > > knowledgeable in virtio.
> >
> > You can test running a big-endian guest (m68k, s390x, sparc64) on your
> > little-endian host.
>
> The linux driver uses le{32,64}_to_cpu for reading fields, if there's 
> any problem then it would be with the host?

Ah right. Since using a BE guest is often simpler for most developers
than accessing a BE host -- in particular to test recent kernel -- I was
hoping this would be an easy enough suggestion.

Anyhow Linux swapping the fields confirms the virtio-sound model has to
do the same.